A spatio-temporal geostatistical approach for sharpening multi-spectral satellite imagery using high-resolution UAV data

Authors: Bo Yang*, University of Central Florida, Timothy Hawthorne, University of Central Florida
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems, Coastal and Marine
Keywords: UAV; spatio-temporal modelling; Cokriging; multi-spectral; seagrass mapping
Session Type: Paper
Day: 4/5/2019
Start / End Time: 9:55 AM / 11:35 AM
Room: Harding, Marriott, Mezzanine Level
Presentation File: No File Uploaded

Seagrasses are one of the most valuable ecosystems on Earth. They provide habitat for marine organisms, protect shorelines from erosion and filter out nutrient pollution. By retrieving the multi-spectral information of seagrass, the high quality data can help us to detect the health condition of seagrass. Therefore, high quality multi-spectral observations with fine spatial resolution and frequent temporal coverage are indispensable in seagrass monitoring and analyses. While UAV imagery usually has high spatial resolution, it only provides limited temporal coverage. Conversely, historical satellite images have regular coverages over a long temporal period, but with relatively lower spatial resolution. This study presents a spatio-temporal Cokriging (ST-Cokriging) method to sharpen satellite data by using a set of time-series multi-spectral satellite imagery and one frame of UAV imagery at high spatial resolution. It extends the traditional Cokriging from the purely spatial domain to the spatio-temporal context. We build a multivariate spatio-temporal statistical model to explicitly describe the spatio-temporal dependence within and between data sets of different spatial and temporal scales. ST-Cokriging algorithm was applied to sharpen Sentinel-2 imagery at 10-m with UAV imagery at 1-m for creating the frequent multi-spectral observations at 1 m spatial resolution in Belize. This ST-cokriging method can effectively handle missing values in input data due to cloud contaminations and other reasons, and generate reliable estimates at both high spatial resolution and high temporal frequency. The uncertainty estimates for the assimilation results are also provided.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login